Symmetry Detection from Real World Images

We call for participation to the first symmetry detection competition from all researchers worldwide, especially students. Three types of symmetries are targeted in real images: reflection, translation and rotation This competition is set-up as a two-stage process, though a participant can enter at any point.

Round One: At ECCV 2010, we organized a full-day tutorial on Computational Symmetry: Past, Present, and Future [link] to launch the competition by providing a succinct and multi-faceted review of computational symmetry in computer vision.

Round Two: We completed the second round of the symmetry detection competition and presented results at a dedicated workshop hosted at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2011 in Colorado Springs, CO.

[Frequently Asked Questions (pdf)]

Image Submission:

We call for submissions of real world images depicting various kinds of symmetries, especially reflection (mirror), rotation and translation (wallpaper) symmetries. We welcome your contribution whether you are a professional or amateur photographer, or simply happen to have some interesting snapshots of symmetries in the real world. Please visit our Flickr group for more information.

[Real World Symmetry Image Submission]

Competition:

Round One
(preliminary, background review, for feedback only):

At ECCV 2010, we organized a full-day tutorial on Computational Symmetry: Past, Present, and Future [link] to launch the competition by providing a succinct and multi-faceted review of computational symmetry in computer vision.

Round Two:

(First countable, CVPR 2011):

Summary:

We completed the second round of the symmetry detection competition and presented results at a dedicated workshop hosted at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2011 in Colorado Springs, CO. The competition was divided into three parts, each focusing on one of three types of symmetries: reflection, rotation and translation respectively. We received eight submissions for symmetry detection, three for reflection, two for rotation and three for translation symmetry. Adding one baseline algorithm to each symmetry group for comparison, we evaluated a total of eleven algorithms.

In comparison, tested algorithms performed best on reflection symmetry, second best on rotation symmetry and last translation symmetry. Looking at sub categories of reflection symmetries, we found that one algorithm performed superior to all other tested algorithms on real images with multiple reflection axis present (judged on recall rate). Another algorithm outperforms all other algorithms on real and synthetic images depicting only a single reflection symmetry, when the number of false positives has been taken into account (judged on precision rate).

We established a testbed for the evaluation of symmetry detection algorithms, devised evaluation metrics and automated the evaluation process. We tested our process on eleven algorithms and established a performance baseline that can be used as reference for future work on symmetry detection.

Workshop:

Completed CVPR 2011 workshop on 'Symmetry Detection from Real World Images'.

We have called for participation in the first symmetry detection competition from all researchers worldwide, especially students. Three types of symmetries are targeted in real images: reflection, translation and rotation. The top winners of Round Two competition have received travel funds from US NSF and presented their algorithms, results and analysis at the Symmetry Detection from Real World Image workshop at CVPR 2011, Colorado Springs, CO, held in June 2011.

Results:

Also see CVPR 2011 workshop.

Reflection

Rotation

Translation

ReflectionRotation.ppt

Tranlsation.ppt

(includes two additional tests on building facades)

Tech Report:

A technical report has been published summarizing method and results of the competition:

TR-CSE-11-012 (3.0MB)

Test Image Sets:

The following are links to archive files including all test images and groundtruth files used in this competition. Please consult the included readme file for further information.

Reflection Symmetry (4.6MB)

Rotation Symmetry (0.4MB)

Translation Symmetry 1 (9.2MB)

Translation Symmetry 2a (Building Facades for Frieze Detection) (6.4MB)

Translation Symmetry 2b (Urban Buildings for Lattice Detection with manual initialization) (1.2MB)

Training Image Sets:

New Training data will be updated after CVPR 2011

Background:

Symmetry is a pervasive phenomenon presenting itself in all forms and scales in natural and manmade environments. Symmetry detection plays an essential role at all levels of human as well as machine perception. Given current interest in the vision community for semantic level tasks like object recognition, scene understanding and activity classification, there remains a missing link in most computer vision systems between a bag of low-level features and high-level semantic meanings. It therefore is acutely important to develop intermediate-level computer vision tools for perceptual-organization and grouping that are robust, general, semantically meaningful and biologically justifiable.

Humans and animals have an innate ability to perceive and take advantage of symmetry in everyday life. In computer vision, however, the use of symmetry in real world images remains inadequate. Even though the pursuit of symmetry detection algorithmically predates computer vision itself and has lasted for almost four decades, few symmetry detection tools are readily available for real world problems. Motivated by a growing intellectual interests of symmetry in vision and graphics (see Figures 1 and 2) and for practical needs of robust computational symmetry tools, this NSF-funded, first of a kind symmetry detection competition will serve as a timely, systematic and quantitative account of state of the art symmetry detection algorithms.

Since the first tutorial on computational symmetry at ACCV 2007 (followed by one at CVPR08 then ECCV10), there has been a drastic jump in the interest in computational symmetry (algorithms and applications), as evidenced by the increasing number of papers submitted and published in both computer vision and computer graphics (Figure 1 from Liu et al 2010 Survey). Several very promising new results have been presented. Most notably, the concept of symmetry has been expanded beyond bilateral reflection symmetry, which dominated computer vision research on symmetry for the past 40 years (Figure 2 from Liu et al 2010 Survey).

ry Publications
Figure 1 Number of Symmetry related publications in major computer vision and computer graphics journals/conferences.

tion Publications
Figure 2 Dividing papers into on reflection symmetry alone versus other types of symmetries (rotation, translation, glide-reflection).

Organizers:

Ingmar Rauschert

Kyle Brocklehurst
Somesh Kashyap

Jingchen Liu

Dr. Yanxi Liu

Advisory Committee:

Jacob Feldman (Rutgers)
Richard Hartley (ANU)
Takeo Kanade (CMU)
Jitendra Malik (U.C. Berkeley)
Doris Schattschneider (Moravian College)
Marjorie Senechal (Smith College)
Christopher Tyler (SKBIC)
Luc Van Gool (ETH Zurich & University of Leuven)
Laurent Younes (Johns Hopkins University)
Alan Yuille (UCLA)
Andrew Zisserman (Oxford)

Resources:

	Training Data and Ground Truth: [Translation] [Rotation] [Reflection]
	Relevant Survey Paper: Computational Symmetry in Computer Vision and Computer Graphics: The publisher can offer the printed copies at $35 (original price $99) + 6 for shipping. Anybody using the online order form can use the code symmCompCVPR and we will honor this price. The online order form is here: https://www.nowpublishers.com/bookorder.aspx?doi=0600000008&product=CGV
	Relevant Databases: Near-Regular Texture Database 2.1D Database from UIUC Zurich Building Image Database

Support:

This competition is supported in part by an NSF grant IIS-1040711 (PI: Y. Liu), and in part by funds from industry sponsors.

News:

Call for Competition: